Insights
AgiBot sets a dance record with 110 Expedition A3 robots
Latest Articles
257 articles
Human Activity Data Collection System Presented for Robot Training
Researchers introduced EgoExplorer, a wearable data collection system and a 100-hour dataset combining first-person video, body motion, finger movements and tactile information. A reconstruction pipeline estimates full-body movements from sparse sensors. Initial tests included a simulated Unitree G1, while physical robot performance and precise object interaction remain unverified.
Generative Bionics plans data factory for humanoid GENE.01
Generative Bionics plans a 5,000-square-meter Physical AI Data Factory in Genoa, with opening scheduled for the fourth quarter of 2027. The facility will train and test GENE.01 humanoids using physical interaction data, simulation and digital twins. It is intended as the first of three planned hubs.
HONOR introduced a compact humanoid robot A1
HONOR presented A1, a compact humanoid prototype focused on mobility. The 136.9 cm robot weighs 40 kg and has a stated maximum speed of 10 km/h. Demonstrations include running, stairs and obstacle avoidance, while its published skill assessment gives manipulation a zero score and navigation three out of five.
Flexible robotic arms learn to predict their movements
Peking University researchers developed a hybrid prediction method for flexible robotic arms. It combines an equilibrium anchor with a learned residual model to limit accumulated motion prediction errors. Tests used a simulated cable-driven arm and a physical soft robotic tail, with improved accuracy and stability over the comparison methods.
Why automated cutting lines are gaining ground in small furniture workshops
Automated cutting is becoming more attractive to small furniture workshops as skilled operators retire and material waste raises costs. The article discusses staffing, unattended operation, cutting optimization and outsourcing, alongside software training, data quality, service coverage and power requirements that workshops need to consider before investing.
Unitree introduced the G1-D wheeled humanoid robot for data collection and AI training
Unitree presented the G1-D, a manipulation-focused version of its G1 platform using a lifting column and optional wheeled mobility. It is aimed at research, education and AI training data collection. Supported tasks include sorting, assembly, dishwasher loading and clothes folding; deployment requires a task-specific safety assessment.
UBTECH Robotics introduced the humanoid robot Walker Tienkung
UBTECH Robotics presented Walker Tienkung, a full-size humanoid for research, education and development. The 170 cm platform weighs 55 kg and emphasizes locomotion, with demonstrations of obstacle avoidance, running and stair climbing. Linux, ROS 2 and optional computing modules provide a software and hardware foundation for experiments.
Tuesday Labs introduced Voxel, a humanoid robot for home use
Tuesday Labs presented Voxel, a home humanoid prototype with a friendly appearance and an Adaptive Operating System. The 150 cm robot weighs 45 kg and has demonstrated obstacle avoidance and goods sorting at an early skill level. Deliveries are planned for 2028, with a $200 refundable reservation and a stated $20,000 product price.
Richtech Robotics introduced the industrial humanoid robot Dex
Richtech Robotics presented Dex, an industrial humanoid prototype with two arms and a choice of mobile or stationary base. The robot uses NVIDIA Jetson Thor computing and interchangeable tooling for sorting and assembly tasks. Its stated price is $60,000, while preliminary skill ratings call for testing against clearly defined requirements.
Google DeepMind trains Apollo humanoid using Gemini 4 Argon
Google DeepMind reported a demonstration in which Gemini 4 Argon analyzed video of a cornhole throw, generated simulation code and transferred the resulting behavior to Apptronik's Apollo humanoid. According to the company, Apollo made seven consecutive successful throws. The report also describes the model's embodied reasoning benchmark result.
Lab-grown muscles allow robots to jump, swim and grab
ETH Zurich researchers jointly designed living muscle tissue and soft hydrogel scaffolds to create bioactuators that jump, swim, move and grasp objects. The roughly 1.5 cm devices show greater movement than earlier systems using comparable muscle tissue. They still depend on nutrient solution, limiting operation outside the laboratory.